Fast simulation of muons produced at the SHiP experiment using Generative Adversarial Networks

Abstract

This paper presents a fast approach to simulating muons produced in interactions of the SPS proton beams with the target of the SHiP experiment. The SHiP experiment will be able to search for new long-lived particles produced in a 400~GeV$/c$ SPS proton beam dump and which travel distances between fifty metres and tens of kilometers. The SHiP detector needs to operate under ultra-low background conditions and requires large simulated samples of muon induced background processes. Through the use of Generative Adversarial Networks it is possible to emulate the simulation of the interaction of 400~GeV$/c$ proton beams with the SHiP target, an otherwise computationally intensive process. For the simulation requirements of the SHiP experiment, generative networks are capable of approximating the full simulation of the dense fixed target, offering a speed increase by a factor of $\mathcal{O}(10^6)$. To evaluate the performance of such an approach, comparisons of the distributions of reconstructed muon momenta in SHiP's spectrometer between samples using the full simulation and samples produced through generative models are presented. The methods discussed in this paper can be generalised and applied to modelling any non-discrete multi-dimensional distribution.

20 pages, 7 figures, amended as per JINST reviewer comments

References in corpus (11)

• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• PYTHIA 6.4 Physics and Manual
• Conditional Generative Adversarial Nets
• Generative Adversarial Networks
• On the Convergence of Adam and Beyond
• A facility to Search for Hidden Particles (SHiP) at the CERN SPS
• Population Based Training of Neural Networks
• A Deep Learning-based Reconstruction of Cosmic Ray-induced Air Showers
• LHC analysis-specific datasets with Generative Adversarial Networks
• Cherenkov Detectors Fast Simulation Using Neural Networks
• Machine learning and multivariate goodness of fit